Update the running production website without service interruption

[jmlord]

Display a warning banner on the website for planned maintenance.

In order to avoid killing long-running tasks, we should have a system that allows:
A. Updating without interruption. For example, any new pipeline run is directed to a new updated instance. When the existing pipeline engine finishes all it's pending tasks, it closes.
B. Running instance finishes all running scripts, without starting new ones. New instance starts with and all the running pipelines are re-launched based on their inputs. Cache system allows to pick up where the previous instance was.
C. Get a count of executing scripts through an API. Poke the API with a cron. When the count reaches 0, perform the update.
D. Kill every running process, this was planned anyways. Launch new instance and re-launch any running pipeline based on inputs. Cache system allows to pick up where the previous instance was, and any killed script re-executes.